[FFmpeg-devel] Anybody has a Core 2? [PATCH] Small SSSE3 optimization
Loren Merritt
lorenm
Sat May 12 21:17:44 CEST 2007
On Sun, 13 May 2007, Zuxy Meng wrote:
> 2007/5/9, Loren Merritt <lorenm at u.washington.edu>:
>>
>> You'd get even more speedup by also using SSE2/xmmregs. And both additions
>> would be better written as macros, no need for code duplication.
>
> I saw your recent contribution in svn HEAD. And most interesting to me
> is that in 'svn log' you gave exact cycles of different routines on
> different CPUs. How did u get that? Are u using a simulator or you
> simply measure using RDTSC?
rdtsc, as per ffmpeg's START_TIMER/STOP_TIMER macros. And then subtract
the time taken by the empty timer, which can be rather large on some
cpus.
For some functions the time is constant. For others (e.g. with random
memory access) it can vary, but is still perfectly reproducible if I run
the same encode twice.
--Loren Merritt
More information about the ffmpeg-devel
mailing list